Graph databases from Rust

Overview

The graph corner is Rust's thinnest storage shelf — no vendor-official Neo4j driver exists, and the JVM's TinkerPop gravity (Java's note) has no Rust pole. What works: neo4rs, the community Bolt-protocol driver Neo4j's own docs point at; gremlin-client for TinkerPop stores; and — the corner's genuine bright spot — oxigraph, an embeddable RDF triple store with SPARQL, written in Rust, usable as a library the way sled or SQLite are. When the graph lives in Postgres (Apache AGE) the relational stack applies instead.

Key points

• neo4rs: async Bolt driver — Graph::new(uri, user, pass), parameterized Cypher via query("…").param("k", v), streamed RowStream results, explicit transactions (graph.start_txn()); community-maintained with Neo4j-team involvement, covering the driver essentials without the official drivers' full session/routing sophistication.
• Cypher discipline transfers verbatim: parameterize always (injection and plan cache both), MATCH shape = index design, transaction functions for retry — the concepts are store-side, see databases & storage systems.
• gremlin-client speaks the TinkerPop protocol (JanusGraph, Neptune's Gremlin endpoint): traversal strings or a typed builder subset; maintained but niche — polyglot shops usually keep Gremlin tooling on the JVM where TinkerPop lives.
• oxigraph flips the model: the store is a crate (also a server binary) — in-process SPARQL 1.1 over RocksDB, standards-faithful, ideal for knowledge-graph-shaped features embedded in a service without operating a separate database.
• Neptune from Rust = openCypher/Gremlin over HTTPS with SigV4 signing (aws-sigv4 + reqwest) — workable, hand-assembled.
• Honest guidance: for a graph-centric product on Neo4j, Rust is a viable client but the JVM ecosystem is a generation richer (OGMs, Spring Data Neo4j); for graph-shaped features inside a Rust service, neo4rs or embedded oxigraph carry their weight — and recursive CTEs in Postgres remain the right first question, as everywhere.

Examples

let graph = Graph::new("bolt://localhost:7687", "neo4j", pass).await?;
let mut rows = graph.execute(
    query("MATCH (p:Person)-[:REPORTS_TO*1..3]->(m:Person {id: $id})
           RETURN p.id AS id, p.name AS name")
        .param("id", mgr_id.to_string())
).await?;
while let Some(row) = rows.next().await? {
    let name: String = row.get("name")?;
    // …
}

Related

• Storage access from Rust — the map — parent map.
• Relational databases from Rust — recursive CTEs: the question to ask before adopting a graph store.
• Databases and other storage systems — when graph models beat relational ones.
• Graph databases from Java and from Go — the counterparts; Java holds the official-driver high ground, Rust holds the embedded-store one.